Summary of The Effect Of Similarity Measures on Accurate Stability Estimates For Local Surrogate Models in Text-based Explainable Ai, by Christopher Burger et al.
The Effect of Similarity Measures on Accurate Stability Estimates for Local Surrogate Models in Text-based Explainable AI
by Christopher Burger, Charles Walter, Thai Le
First submitted to arxiv on: 22 Jun 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Cryptography and Security (cs.CR)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper explores the vulnerability of local surrogate methods used for explainable AI (XAI) to adversarial perturbations on input data. The authors identify that a poor choice of similarity measure can lead to inaccurate conclusions about the efficacy of XAI methods, highlighting the need for a robust and reliable measure. They investigate various text-based ranked list measures, including Kendall’s Tau, Spearman’s Footrule, and Rank-biased Overlap, to determine how different measures and thresholds affect the results of common adversarial attack processes. The study finds that certain measures are overly sensitive, leading to incorrect estimates of stability. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper looks at how well local surrogate methods work when they’re attacked with fake data that makes the explanation change but not the original input’s meaning. It shows that many methods have weaknesses, but we don’t really know why. One important part is choosing the right way to measure how different explanations are from each other. If you pick a bad one, you might get wrong ideas about which XAI method works best. The researchers try out different ways to compare ranked lists and find that some are too sensitive and give false results. |
